This paper describes work in progress at Oxford Brookes University. The aim of the project is to define a higher-level diagram description language for the World Wide Web, which captures the structure and the semantics of a diagram and enables the generation of accessible presentations in different modalities such as speech, text, graphic, etc. This project, called Graphical Structure Semantic Markup Languages (GraSSML), is decomposed into three levels: presentation, structure and semantics. Each of these levels captures a specific aspect of a diagram. The language at the structure level (called "ZineML") uses SVG as the output renderer at the presentation level. The semantic level language is highly dependent on the type of diagram considered and the field in which it is used. The paper outlines the relevant limitation of SVG and describes our approach which addresses some of these limitations and can allow web graphics to be written easily and made accessible.